Horizon CDT Research Highlights

Research Highlights

Multi-Modal Data Fusion Framework for Classifying Heavy Goods Vehicle Driving Behaviour

  Jimiama Mafeni Mase (2018 cohort)   www.linkedin.com/in/jimiama-mase


The aim of this research is to investigate data fusion and mining techniques for analysing and merging different  Heavy Goods Vehicle (HGV) driving data (e.g. telematics, images, expert knowledge). We hypothesise that fusing heterogeneous data sources representing different facets of driving behaviour, will provide a more accurate, reliable and holistic description of a driver's behaviour. Determining driving styles and the factors causing incidents or accidents in real time can assist stakeholders to promote actions and develop feedback systems to reduce risks, costs and increase safety in roads.


Heavy Goods Vehicles (HGVs) are at the forefront of trade and commerce in the United Kingdom. Both private and public sectors rely on HGVs road transport for the delivery of goods and services. For instance, 1.4 billion tonnes of freight were transported by road between 2016 and 2017 in the UK. Over the same period, a total of 7.8 million tonnes of freight was moved to or from the UK by HGVs [1]. As a result of the importance of HGVs to a nation’s economy, there are great efforts being employed to reduce the number of road accidents caused by HGVs as well as the costs associated with HGVs such as fuel and maintenance costs. These issues are due to one or more of the following factors: vehicle characteristics, weather conditions, company policies and driver behaviour, with driver behaviour being by far the leading determinant [3].

In psychology, risky or dangerous driving behaviour is classified into intentional (deliberate) and unintentional behaviours [2]. The intentional behaviours consist of violations and mistakes which involve inappropriate actions while unintentional behaviours consists of slips and lapses which involve errors due to memory or attention failures [4]. The intentional behaviours can be captured from vehicle operation data (Telematics) such as over speeding, harsh braking, harsh cornering etc. However, capturing the unintentional behaviours is very complex because they can’t be directly measured and need to be extracted from other measurements such as physiological measures and images. A greater constraint exists in fusing and analysing these heterogeneous data sources in order to obtain a holistic view of risky driver behaviour.

Academic Contributions

The contributions of this research will regard novel algorithms or frameworks to tackle Data fusion of multi-level data sources and classifying HGV driving behaviour.

Research Questions

This research plans on answering the following research questions:

  1. How can we detect and classify risky HGV driving behaviours using the different data sources?
  2. What are the effects of the technologies used to capture the driving data on their driving behaviours?
  3. How can we combine the classifications from the different data sources into one unified output/outcome?

We also have sub-questions which will assist in answering the main research questions.:

  1. How can we develop accurate and reliable detection models for each of the data sources?
  2. How can we develop a trustworthy and interpretable fusion framework?
  3. How can we develop a computationally efficient fusion framework?


  1. Department of Transport: International and Domestic Road Freight Statistics, United Kingdom 2017 (https://www.gov.uk/government/statistics/road-freight-statistics-2017)
  2. Jafarpour, S. and Rahimi-Movaghar, V., 2014. Determinants of risky driving behavior: a narrative review. Medical journal of the Islamic Republic of Iran, 28, p.142.
  3. WHO. World report on road traffic injury prevention. Geneva: World Health Organization; 2015.
  4. Reason, J., 1990. Human error. Cambridge university press.
  5. Dula, C.S. and Geller, E.S., 2003. Risky, aggressive, or emotional driving: Addressing the need for consistent communication in research. Journal of safety research, 34(5), pp.559-566.


This author is supported by the Horizon Centre for Doctoral Training at the University of Nottingham (RCUK Grant No. EP/L015463/1) and Microlise.